Poll
Score of Milov vs. Rybka handicap match? (Closed)
| Milov wins by 7-1 or more | 2 | 2% | |
| Milovwins by 6.5-1.5 | 1 | 1% | |
| Milov wins by 6-2 | 0 | 0% | |
| Milov wins by 5.5-2.5 | 3 | 3% | |
| Milov wins by 5-3 | 7 | 7% | |
| Milov wins by 4.5-3.5 | 10 | 10% | |
| Tie match 4-4 | 11 | 11% | |
| Rybka wins by 4.5-3.5 | 18 | 18% | |
| Rybka wins by 5-3 | 16 | 16% | |
| Rybka wins by 5.5-2.5 | 10 | 10% | |
| Rybka wins by 6-2 | 12 | 12% | |
| Rybka wins by 6.5-1.5 | 4 | 4% | |
| Rybka wins by 7-1 or more | 8 | 8% |
![[us]](/mwf/v12/flags/us.png "United States")
It is a bit surprising that the handicap value jumps so much between 12 and 13 plies.
Clearly, the function f(n) which measures the elo value of a handicap as a function of a ply level n, gets to infinity (since the position must be a tablebase win for somebody shuffling 30-piece tb). But since 900 ELO gap already means only one draw in 100 games, f(n) < 1000 for all practical purposes, even for large n. I would expect the growth of this function to be quite slow, linear at most.
What would be your guess, at what ply level will the handicap value actually get to 800 or so? In other words, at what depth will Rybka start scoring 99+% with this handicap against itself?
Clearly, the function f(n) which measures the elo value of a handicap as a function of a ply level n, gets to infinity (since the position must be a tablebase win for somebody shuffling 30-piece tb). But since 900 ELO gap already means only one draw in 100 games, f(n) < 1000 for all practical purposes, even for large n. I would expect the growth of this function to be quite slow, linear at most.
What would be your guess, at what ply level will the handicap value actually get to 800 or so? In other words, at what depth will Rybka start scoring 99+% with this handicap against itself?
It's probably just due to small sample size at 13 plies. I think the level at which Rybka would score 99+% at this handicap is beyond anything that is practical to test. It might happen at one day per move or one week per move or even one century per move. So far no program has the sense to play conservatively (avoid unclear complications) when up in material (as far as I know). Maybe Rybka 4 might address this problem.
Here is something for you amusement - after Rybka 3sp-w32 played Gaia (CCRL 2500) about even in a knight odds match, I let them play a short match at a somewhat longer control (8+2") at "rook vs f7 pawn" handicap. The results are once again about even: 9.5:10.5 (+7=5-8) Rybka loss. I glanced at Rybka wins, and they look very impressive to me, most of them came in late middlegame, and it is amazing to see Rybka playing normal chess a rook down.
Here is a typical game, quite pretty, and also a nice endgame win with extra R+B totally helpless agains 3 white pawns.
Based on what I see so far, I am surprised it looked so helpless with knight handicap against Meyer. Do you think a current version would fare a bit better, for example avoiding material reduction?
[Event "[#47] Rybka 3 1-cpu w32 - gaia"]
[Site "Local computer"]
[Date "2008.08.10"]
[Round "2"]
[White "Rybka 3 1-cpu w32"]
[Black "gaia32"]
[Result "1-0"]
[FEN "rnbqkbnr/ppppp1pp/8/8/8/P7/1PPPPPPP/1NBQKBNR w Kkq - 0 1"]
[SetUp "1"]
[TimeControl "8 min/game + 2 sec/move"]
{[%t Long] White checkmates.}{[%t bLon] Processor: Genuine Intel(R) CPU
T2080 @ 1.73GHz^13 ^10 White: Rybka 3 1-cpu w32, hash size: 32M, opening book:
HandicapBook^13 ^10 Black: gaia32, hash size: 32M, opening book: NarrowBook^13
^10 } 1. d4 {[%clk 0:07:48][%clko 0:08:00][%emt 0:00:12]}{[%eval -251,12]}{[%t
bLon] White out of book}{Black out of book} 1... Nf6 {[%clk 0:07:36][%clko
0:07:48][%emt 0:00:23]}{[%eval -335,12]} 2. Nc3 {[%clk 0:07:29][%clko
0:07:36][%emt 0:00:20]}{[%eval -263,12]} 2... d5 {[%clk 0:07:16][%clko
0:07:29][%emt 0:00:22]}{[%eval -338,11]} 3. Nf3 {[%clk 0:06:57][%clko
0:07:16][%emt 0:00:34]}{[%eval -274,12]} 3... g6 {[%clk 0:06:56][%clko
0:06:57][%emt 0:00:22]}{[%eval -341,12]} 4. Bf4 {[%clk 0:06:51][%clko
0:06:56][%emt 0:00:08]}{[%eval -239,11]} 4... Bg7 {[%clk 0:06:37][%clko
0:06:51][%emt 0:00:21]}{[%eval -343,12]} 5. e3 {[%clk 0:06:47][%clko
0:06:37][%emt 0:00:06]}{[%eval -247,11]} 5... O-O {[%clk 0:06:19][%clko
0:06:47][%emt 0:00:20]}{[%eval -342,11]} 6. Be2 {[%clk 0:06:39][%clko
0:06:19][%emt 0:00:10]}{[%eval -253,11]} 6... Nh5 {[%clk 0:06:03][%clko
0:06:39][%emt 0:00:19]}{[%eval -344,12]} 7. Be5 {[%clk 0:06:26][%clko
0:06:03][%emt 0:00:15]}{[%eval -263,12]} 7... Nc6 {[%clk 0:05:46][%clko
0:06:26][%emt 0:00:18]}{[%eval -341,12]} 8. h3 {[%clk 0:06:19][%clko
0:05:46][%emt 0:00:09]}{[%eval -263,11]} 8... a6 {[%clk 0:05:31][%clko
0:06:19][%emt 0:00:17]}{[%eval -349,11]} 9. Bh2 {[%clk 0:06:07][%clko
0:05:31][%emt 0:00:14]}{[%eval -245,12]} 9... e6 {[%clk 0:05:16][%clko
0:06:07][%emt 0:00:17]}{[%eval -358,11]} 10. Qd3 {[%clk 0:06:00][%clko
0:05:16][%emt 0:00:09]}{[%eval -248,10]} 10... b5 {[%clk 0:05:02][%clko
0:06:00][%emt 0:00:16]}{[%eval -366,11]} 11. O-O {[%clk 0:05:56][%clko
0:05:02][%emt 0:00:07]}{[%eval -244,11]} 11... Bb7 {[%clk 0:04:49][%clko
0:05:56][%emt 0:00:15]}{[%eval -373,11]} 12. b4 {[%clk 0:05:49][%clko
0:04:49][%emt 0:00:08]}{[%eval -246,10]} 12... Rc8 {[%clk 0:04:36][%clko
0:05:49][%emt 0:00:15]}{[%eval -375,11]} 13. Kh1 {[%clk 0:05:34][%clko
0:04:36][%emt 0:00:17]}{[%eval -244,10]} 13... Bf6 {[%clk 0:04:23][%clko
0:05:34][%emt 0:00:14]}{[%eval -382,11]} 14. Rd1 {[%clk 0:05:26][%clko
0:04:23][%emt 0:00:10]}{[%eval -244,10]} 14... Be7 {[%clk 0:04:12][%clko
0:05:26][%emt 0:00:14]}{[%eval -377,11]} 15. e4 {[%clk 0:05:15][%clko
0:04:12][%emt 0:00:13]}{[%eval -247,11]} 15... Nb8 {[%clk 0:04:00][%clko
0:05:15][%emt 0:00:13]}{[%eval -376,10]} 16. exd5 {[%clk 0:05:08][%clko
0:04:00][%emt 0:00:08]}{[%eval -230,10]} 16... exd5 {[%clk 0:03:50][%clko
0:05:08][%emt 0:00:13]}{[%eval -387,12]} 17. Ne5 {[%clk 0:04:49][%clko
0:03:50][%emt 0:00:21]}{[%eval -248,11]} 17... Ng7 {[%clk 0:03:39][%clko
0:04:49][%emt 0:00:12]}{[%eval -383,11]} 18. Bf3 {[%clk 0:04:35][%clko
0:03:39][%emt 0:00:17]}{[%eval -244,11]} 18... Nf5 {[%clk 0:03:30][%clko
0:04:35][%emt 0:00:12]}{[%eval -389,11]} 19. Re1 {[%clk 0:04:32][%clko
0:03:30][%emt 0:00:05]}{[%eval -238,11]} 19... c6 {[%clk 0:03:20][%clko
0:04:32][%emt 0:00:11]}{[%eval -387,11]} 20. Nd1 {[%clk 0:04:11][%clko
0:03:20][%emt 0:00:22]}{[%eval -248,11]} 20... Nd7 {[%clk 0:03:11][%clko
0:04:11][%emt 0:00:11]}{[%eval -382,11]} 21. Bg4 {[%clk 0:04:02][%clko
0:03:11][%emt 0:00:11]}{[%eval -269,11]} 21... Bh4 {[%clk 0:03:03][%clko
0:04:02][%emt 0:00:10]}{[%eval -380,10]} 22. Ne3 {[%clk 0:03:51][%clko
0:03:03][%emt 0:00:13]}{[%eval -221,11]} 22... Nxe5 {[%clk 0:02:55][%clko
0:03:51][%emt 0:00:10]}{[%eval -346,12]} 23. Bxe5 {[%clk 0:03:53][%clko
0:02:55]}{[%eval -221,10]} 23... Qd7 {[%clk 0:02:47][%clko 0:03:53][%emt
0:00:10]}{[%eval -350,11]} 24. g3 {[%clk 0:03:45][%clko 0:02:47][%emt 0:00:10]}
{[%eval -204,11]} 24... Be7 {[%clk 0:02:39][%clko 0:03:45][%emt 0:00:09]}
{[%eval -367,10]} 25. Bxf5 {[%clk 0:03:36][%clko 0:02:39][%emt 0:00:11]}{[%eval
-209,11]} 25... gxf5 {[%clk 0:02:32][%clko 0:03:36][%emt 0:00:09]}{[%eval
-264,12]} 26. g4 {[%clk 0:03:26][%clko 0:02:32][%emt 0:00:11]}{[%eval -155,11]}
26... Rf7 {[%clk 0:02:26][%clko 0:03:26][%emt 0:00:09]}{[%eval -280,12]} 27.
Nxf5 {[%clk 0:03:22][%clko 0:02:26][%emt 0:00:06]}{[%eval -142,10]} 27... Bf8
{[%clk 0:02:19][%clko 0:03:22][%emt 0:00:08]}{[%eval -280,11]} 28. Qe3 {[%clk
0:03:17][%clko 0:02:19][%emt 0:00:07]}{[%eval -151,11]} 28... Qd8 {[%clk
0:02:13][%clko 0:03:17][%emt 0:00:08]}{[%eval -278,11]} 29. Qf4 {[%clk
0:03:03][%clko 0:02:13][%emt 0:00:16]}{[%eval -114,10]} 29... Ra8 {[%clk
0:02:07][%clko 0:03:03][%emt 0:00:08]}{[%eval -268,12]} 30. Re3 {[%clk
0:02:58][%clko 0:02:07][%emt 0:00:06]}{[%eval -110,12]} 30... Bc8 {[%clk
0:02:01][%clko 0:02:58][%emt 0:00:08]}{[%eval -264,13]} 31. Nh6+ {[%clk
0:02:53][%clko 0:02:01][%emt 0:00:08]}{[%eval -100,11]} 31... Bxh6 {[%clk
0:01:55][%clko 0:02:53][%emt 0:00:07]}{[%eval -238,13]} 32. Qxh6 {[%clk
0:02:35][%clko 0:01:55][%emt 0:00:20]}{[%eval -110,13]} 32... Bd7 {[%clk
0:01:51][%clko 0:02:35][%emt 0:00:07]}{[%eval -260,13]} 33. Kg2 {[%clk
0:02:12][%clko 0:01:51][%emt 0:00:25]}{[%eval -95,13]} 33... Rc8 {[%clk
0:01:46][%clko 0:02:12][%emt 0:00:07]}{[%eval -235,11]} 34. Kg3 {[%clk
0:02:05][%clko 0:01:46][%emt 0:00:09]}{[%eval -95,12]} 34... a5 {[%clk
0:01:41][%clko 0:02:05][%emt 0:00:07]}{[%eval -299,12]} 35. bxa5 {[%clk
0:02:03][%clko 0:01:41][%emt 0:00:04]}{[%eval -78,10]} 35... c5 {[%clk
0:01:37][%clko 0:02:03][%emt 0:00:06]}{[%eval -291,11]} 36. dxc5 {[%clk
0:01:58][%clko 0:01:37][%emt 0:00:06]}{[%eval -72,11]} 36... Rxc5 {[%clk
0:01:33][%clko 0:01:58][%emt 0:00:06]}{[%eval -295,11]} 37. Qb6 {[%clk
0:01:50][%clko 0:01:33][%emt 0:00:12]}{[%eval -81,11]} 37... Qc8 {[%clk
0:01:29][%clko 0:01:50][%emt 0:00:06]}{[%eval -317,11]} 38. f4 {[%clk
0:01:24][%clko 0:01:29][%emt 0:00:27]}{[%eval -89,10]} 38... Rxc2 {[%clk
0:01:24][%clko 0:01:24][%emt 0:00:06]}{[%eval -362,10]} 39. Rd3 {[%clk
0:01:16][%clko 0:01:24][%emt 0:00:11]}{[%eval -92,10]} 39... Qf8 {[%clk
0:01:21][%clko 0:01:16][%emt 0:00:06]}{[%eval -377,9]} 40. a6 {[%clk
0:01:14][%clko 0:01:21][%emt 0:00:04]}{[%eval -74,9]} 40... Qc8 {[%clk
0:01:17][%clko 0:01:14][%emt 0:00:06]}{[%eval -302,9]} 41. a7 {[%clk
0:01:13][%clko 0:01:17][%emt 0:00:03]}{[%eval -18,9]} 41... Qa8 {[%clk
0:01:14][%clko 0:01:13][%emt 0:00:05]}{[%eval -291,10]} 42. Rc3 {[%clk
0:01:11][%clko 0:01:14][%emt 0:00:04]}{[%eval 20,9]} 42... Rxc3+ {[%clk
0:01:11][%clko 0:01:11][%emt 0:00:05]}{[%eval -110,11]} 43. Bxc3 {[%clk
0:01:13][%clko 0:01:11]}{[%eval 20,7]} 43... Re7 {[%clk 0:01:08][%clko
0:01:13][%emt 0:00:05]}{[%eval -79,11]} 44. Qd4 {[%clk 0:01:11][%clko
0:01:08][%emt 0:00:03]}{[%eval 22,10]} 44... Bc6 {[%clk 0:01:05][%clko
0:01:11][%emt 0:00:05]}{[%eval -60,11]} 45. Qh8+ {[%clk 0:01:05][%clko
0:01:05][%emt 0:00:08]}{[%eval 8,9]} 45... Kf7 {[%clk 0:01:02][%clko
0:01:05][%emt 0:00:05]}{[%eval -86,11]} 46. Qxh7+ {[%clk 0:00:55][%clko
0:01:02][%emt 0:00:13]}{[%eval 1,11]} 46... Ke8 {[%clk 0:00:59][%clko
0:00:55][%emt 0:00:05]}{[%eval -83,12]} 47. Qg6+ {[%clk 0:00:50][%clko
0:00:59][%emt 0:00:06]}{[%eval 16,11]} 47... Kd7 {[%clk 0:00:57][%clko
0:00:50][%emt 0:00:04]}{[%eval 0,12]} 48. Be5 {[%clk 0:00:49][%clko
0:00:57][%emt 0:00:03]}{[%eval 16,11]} 48... Kc8 {[%clk 0:00:54][%clko
0:00:49][%emt 0:00:04]}{[%eval -79,11]} 49. Qg8+ {[%clk 0:00:37][%clko
0:00:54][%emt 0:00:14]}{[%eval 3,11]} 49... Be8 {[%clk 0:00:52][%clko
0:00:37][%emt 0:00:04]}{[%eval -65,11]} 50. Bd4 {[%clk 0:00:37][%clko
0:00:52][%emt 0:00:02]}{[%eval 3,11]} 50... Qc6 {[%clk 0:00:49][%clko
0:00:37][%emt 0:00:04]}{[%eval 0,11]} 51. Qf8 {[%clk 0:00:37][%clko
0:00:49][%emt 0:00:02]}{[%eval 63,8]} 51... Kd7 {[%clk 0:00:47][%clko
0:00:37][%emt 0:00:04]}{[%eval -20,10]} 52. g5 {[%clk 0:00:38][%clko
0:00:47][%emt 0:00:01]}{[%eval 92,7]} 52... Re4 {[%clk 0:00:45][%clko
0:00:38][%emt 0:00:04]}{[%eval -70,9]} 53. h4 {[%clk 0:00:28][%clko
0:00:45][%emt 0:00:11]}{[%eval 125,8]} 53... Kc7 {[%clk 0:00:43][%clko
0:00:28][%emt 0:00:04]}{[%eval -4,9]} 54. h5 {[%clk 0:00:27][%clko
0:00:43][%emt 0:00:05]}{[%eval 157,7]} 54... Kb7 {[%clk 0:00:41][%clko
0:00:27][%emt 0:00:04]}{[%eval 0,9]} 55. g6 {[%clk 0:00:23][%clko 0:00:41][%emt
0:00:05]}{[%eval 209,8]} 55... Qe6 {[%clk 0:00:40][%clko 0:00:23][%emt
0:00:04]}{[%eval 0,9]} 56. f5 {[%clk 0:00:19][%clko 0:00:40][%emt 0:00:06]}
{[%eval 238,9]} 56... Qc6 {[%clk 0:00:38][%clko 0:00:19][%emt 0:00:04]}{[%eval
-84,8]} 57. Qc5 {[%clk 0:00:18][%clko 0:00:38][%emt 0:00:02]}{[%eval 309,8]}
57... Qxc5 {[%clk 0:00:36][%clko 0:00:18][%emt 0:00:04]}{[%eval -103,10]} 58.
Bxc5 {[%clk 0:00:19][%clko 0:00:36][%emt 0:00:01]}{[%eval 329,10]} 58... Re5
{[%clk 0:00:35][%clko 0:00:19][%emt 0:00:04]}{[%eval 0,10]} 59. Kf4 {[%clk
0:00:19][%clko 0:00:35][%emt 0:00:02]}{[%eval 388,9]} 59... Re4+ {[%clk
0:00:33][%clko 0:00:19][%emt 0:00:03]}{[%eval 0,12]} 60. Kg5 {[%clk
0:00:21][%clko 0:00:33]}{[%eval 388,8]} 60... d4 {[%clk 0:00:32][%clko
0:00:21][%emt 0:00:03]}{[%eval -19,10]} 61. Bxd4 {[%clk 0:00:17][%clko
0:00:32][%emt 0:00:06]}{[%eval 448,10]} 61... Rxd4 {[%clk 0:00:30][%clko
0:00:17][%emt 0:00:03]}{[%eval 0,11]} 62. h6 {[%clk 0:00:14][%clko
0:00:30][%emt 0:00:04]}{[%eval 512,9]} 62... Rd8 {[%clk 0:00:28][%clko
0:00:14][%emt 0:00:03]}{[%eval 418,10]} 63. g7 {[%clk 0:00:14][%clko
0:00:28][%emt 0:00:02]}{[%eval 733,8]} 63... Bf7 {[%clk 0:00:28][%clko
0:00:14][%emt 0:00:03]}{[%eval 423,12]} 64. h7 {[%clk 0:00:12][%clko
0:00:28][%emt 0:00:04]}{[%eval 733,7]} 64... Rg8 {[%clk 0:00:27][%clko
0:00:12][%emt 0:00:03]}{[%eval 423,12]} 65. h8=Q {[%clk 0:00:10][%clko
0:00:27][%emt 0:00:03]}{[%eval 793,7]} 65... Kxa7 {[%clk 0:00:25][%clko
0:00:10][%emt 0:00:03]}{[%eval 537,11]} 66. Kf6 {[%clk 0:00:10][%clko
0:00:25][%emt 0:00:03]}{[%eval 790,8]} 66... Bd5 {[%clk 0:00:24][%clko
0:00:10][%emt 0:00:03]}{[%eval 654,11]} 67. Ke7 {[%clk 0:00:10][%clko
0:00:24][%emt 0:00:02]}{[%eval 850,7]} 67... b4 {[%clk 0:00:23][%clko
0:00:10][%emt 0:00:03]}{[%eval 826,11]} 68. axb4 {[%clk 0:00:10][%clko
0:00:23][%emt 0:00:02]}{[%eval 916,8]} 68... Kb7 {[%clk 0:00:21][%clko
0:00:10][%emt 0:00:03]}{[%eval 1143,11]} 69. f6 {[%clk 0:00:10][%clko
0:00:21][%emt 0:00:02]}{[%eval 1187,6]} 69... Kc6 {[%clk 0:00:21][%clko
0:00:10][%emt 0:00:03]}{[%eval 1950,11]} 70. Qxg8 {[%clk 0:00:10][%clko
0:00:21][%emt 0:00:02]}{[%eval 1981,6]} 70... Bxg8 {[%clk 0:00:21][%clko
0:00:10][%emt 0:00:03]}{[%eval 29965,11]} 71. f7 {[%clk 0:00:09][%clko
0:00:21][%emt 0:00:02]}{[%eval 1983,6]} 71... Bxf7 {[%clk 0:00:20][%clko
0:00:09][%emt 0:00:03]}{[%eval 29986,12]} 72. Kxf7 {[%clk 0:00:09][%clko
0:00:20][%emt 0:00:02]}{[%eval 3267,9]} 1-0
Here is a typical game, quite pretty, and also a nice endgame win with extra R+B totally helpless agains 3 white pawns.
Based on what I see so far, I am surprised it looked so helpless with knight handicap against Meyer. Do you think a current version would fare a bit better, for example avoiding material reduction?
[Event "[#47] Rybka 3 1-cpu w32 - gaia"]
[Site "Local computer"]
[Date "2008.08.10"]
[Round "2"]
[White "Rybka 3 1-cpu w32"]
[Black "gaia32"]
[Result "1-0"]
[FEN "rnbqkbnr/ppppp1pp/8/8/8/P7/1PPPPPPP/1NBQKBNR w Kkq - 0 1"]
[SetUp "1"]
[TimeControl "8 min/game + 2 sec/move"]
{[%t Long] White checkmates.}{[%t bLon] Processor: Genuine Intel(R) CPU
T2080 @ 1.73GHz^13 ^10 White: Rybka 3 1-cpu w32, hash size: 32M, opening book:
HandicapBook^13 ^10 Black: gaia32, hash size: 32M, opening book: NarrowBook^13
^10 } 1. d4 {[%clk 0:07:48][%clko 0:08:00][%emt 0:00:12]}{[%eval -251,12]}{[%t
bLon] White out of book}{Black out of book} 1... Nf6 {[%clk 0:07:36][%clko
0:07:48][%emt 0:00:23]}{[%eval -335,12]} 2. Nc3 {[%clk 0:07:29][%clko
0:07:36][%emt 0:00:20]}{[%eval -263,12]} 2... d5 {[%clk 0:07:16][%clko
0:07:29][%emt 0:00:22]}{[%eval -338,11]} 3. Nf3 {[%clk 0:06:57][%clko
0:07:16][%emt 0:00:34]}{[%eval -274,12]} 3... g6 {[%clk 0:06:56][%clko
0:06:57][%emt 0:00:22]}{[%eval -341,12]} 4. Bf4 {[%clk 0:06:51][%clko
0:06:56][%emt 0:00:08]}{[%eval -239,11]} 4... Bg7 {[%clk 0:06:37][%clko
0:06:51][%emt 0:00:21]}{[%eval -343,12]} 5. e3 {[%clk 0:06:47][%clko
0:06:37][%emt 0:00:06]}{[%eval -247,11]} 5... O-O {[%clk 0:06:19][%clko
0:06:47][%emt 0:00:20]}{[%eval -342,11]} 6. Be2 {[%clk 0:06:39][%clko
0:06:19][%emt 0:00:10]}{[%eval -253,11]} 6... Nh5 {[%clk 0:06:03][%clko
0:06:39][%emt 0:00:19]}{[%eval -344,12]} 7. Be5 {[%clk 0:06:26][%clko
0:06:03][%emt 0:00:15]}{[%eval -263,12]} 7... Nc6 {[%clk 0:05:46][%clko
0:06:26][%emt 0:00:18]}{[%eval -341,12]} 8. h3 {[%clk 0:06:19][%clko
0:05:46][%emt 0:00:09]}{[%eval -263,11]} 8... a6 {[%clk 0:05:31][%clko
0:06:19][%emt 0:00:17]}{[%eval -349,11]} 9. Bh2 {[%clk 0:06:07][%clko
0:05:31][%emt 0:00:14]}{[%eval -245,12]} 9... e6 {[%clk 0:05:16][%clko
0:06:07][%emt 0:00:17]}{[%eval -358,11]} 10. Qd3 {[%clk 0:06:00][%clko
0:05:16][%emt 0:00:09]}{[%eval -248,10]} 10... b5 {[%clk 0:05:02][%clko
0:06:00][%emt 0:00:16]}{[%eval -366,11]} 11. O-O {[%clk 0:05:56][%clko
0:05:02][%emt 0:00:07]}{[%eval -244,11]} 11... Bb7 {[%clk 0:04:49][%clko
0:05:56][%emt 0:00:15]}{[%eval -373,11]} 12. b4 {[%clk 0:05:49][%clko
0:04:49][%emt 0:00:08]}{[%eval -246,10]} 12... Rc8 {[%clk 0:04:36][%clko
0:05:49][%emt 0:00:15]}{[%eval -375,11]} 13. Kh1 {[%clk 0:05:34][%clko
0:04:36][%emt 0:00:17]}{[%eval -244,10]} 13... Bf6 {[%clk 0:04:23][%clko
0:05:34][%emt 0:00:14]}{[%eval -382,11]} 14. Rd1 {[%clk 0:05:26][%clko
0:04:23][%emt 0:00:10]}{[%eval -244,10]} 14... Be7 {[%clk 0:04:12][%clko
0:05:26][%emt 0:00:14]}{[%eval -377,11]} 15. e4 {[%clk 0:05:15][%clko
0:04:12][%emt 0:00:13]}{[%eval -247,11]} 15... Nb8 {[%clk 0:04:00][%clko
0:05:15][%emt 0:00:13]}{[%eval -376,10]} 16. exd5 {[%clk 0:05:08][%clko
0:04:00][%emt 0:00:08]}{[%eval -230,10]} 16... exd5 {[%clk 0:03:50][%clko
0:05:08][%emt 0:00:13]}{[%eval -387,12]} 17. Ne5 {[%clk 0:04:49][%clko
0:03:50][%emt 0:00:21]}{[%eval -248,11]} 17... Ng7 {[%clk 0:03:39][%clko
0:04:49][%emt 0:00:12]}{[%eval -383,11]} 18. Bf3 {[%clk 0:04:35][%clko
0:03:39][%emt 0:00:17]}{[%eval -244,11]} 18... Nf5 {[%clk 0:03:30][%clko
0:04:35][%emt 0:00:12]}{[%eval -389,11]} 19. Re1 {[%clk 0:04:32][%clko
0:03:30][%emt 0:00:05]}{[%eval -238,11]} 19... c6 {[%clk 0:03:20][%clko
0:04:32][%emt 0:00:11]}{[%eval -387,11]} 20. Nd1 {[%clk 0:04:11][%clko
0:03:20][%emt 0:00:22]}{[%eval -248,11]} 20... Nd7 {[%clk 0:03:11][%clko
0:04:11][%emt 0:00:11]}{[%eval -382,11]} 21. Bg4 {[%clk 0:04:02][%clko
0:03:11][%emt 0:00:11]}{[%eval -269,11]} 21... Bh4 {[%clk 0:03:03][%clko
0:04:02][%emt 0:00:10]}{[%eval -380,10]} 22. Ne3 {[%clk 0:03:51][%clko
0:03:03][%emt 0:00:13]}{[%eval -221,11]} 22... Nxe5 {[%clk 0:02:55][%clko
0:03:51][%emt 0:00:10]}{[%eval -346,12]} 23. Bxe5 {[%clk 0:03:53][%clko
0:02:55]}{[%eval -221,10]} 23... Qd7 {[%clk 0:02:47][%clko 0:03:53][%emt
0:00:10]}{[%eval -350,11]} 24. g3 {[%clk 0:03:45][%clko 0:02:47][%emt 0:00:10]}
{[%eval -204,11]} 24... Be7 {[%clk 0:02:39][%clko 0:03:45][%emt 0:00:09]}
{[%eval -367,10]} 25. Bxf5 {[%clk 0:03:36][%clko 0:02:39][%emt 0:00:11]}{[%eval
-209,11]} 25... gxf5 {[%clk 0:02:32][%clko 0:03:36][%emt 0:00:09]}{[%eval
-264,12]} 26. g4 {[%clk 0:03:26][%clko 0:02:32][%emt 0:00:11]}{[%eval -155,11]}
26... Rf7 {[%clk 0:02:26][%clko 0:03:26][%emt 0:00:09]}{[%eval -280,12]} 27.
Nxf5 {[%clk 0:03:22][%clko 0:02:26][%emt 0:00:06]}{[%eval -142,10]} 27... Bf8
{[%clk 0:02:19][%clko 0:03:22][%emt 0:00:08]}{[%eval -280,11]} 28. Qe3 {[%clk
0:03:17][%clko 0:02:19][%emt 0:00:07]}{[%eval -151,11]} 28... Qd8 {[%clk
0:02:13][%clko 0:03:17][%emt 0:00:08]}{[%eval -278,11]} 29. Qf4 {[%clk
0:03:03][%clko 0:02:13][%emt 0:00:16]}{[%eval -114,10]} 29... Ra8 {[%clk
0:02:07][%clko 0:03:03][%emt 0:00:08]}{[%eval -268,12]} 30. Re3 {[%clk
0:02:58][%clko 0:02:07][%emt 0:00:06]}{[%eval -110,12]} 30... Bc8 {[%clk
0:02:01][%clko 0:02:58][%emt 0:00:08]}{[%eval -264,13]} 31. Nh6+ {[%clk
0:02:53][%clko 0:02:01][%emt 0:00:08]}{[%eval -100,11]} 31... Bxh6 {[%clk
0:01:55][%clko 0:02:53][%emt 0:00:07]}{[%eval -238,13]} 32. Qxh6 {[%clk
0:02:35][%clko 0:01:55][%emt 0:00:20]}{[%eval -110,13]} 32... Bd7 {[%clk
0:01:51][%clko 0:02:35][%emt 0:00:07]}{[%eval -260,13]} 33. Kg2 {[%clk
0:02:12][%clko 0:01:51][%emt 0:00:25]}{[%eval -95,13]} 33... Rc8 {[%clk
0:01:46][%clko 0:02:12][%emt 0:00:07]}{[%eval -235,11]} 34. Kg3 {[%clk
0:02:05][%clko 0:01:46][%emt 0:00:09]}{[%eval -95,12]} 34... a5 {[%clk
0:01:41][%clko 0:02:05][%emt 0:00:07]}{[%eval -299,12]} 35. bxa5 {[%clk
0:02:03][%clko 0:01:41][%emt 0:00:04]}{[%eval -78,10]} 35... c5 {[%clk
0:01:37][%clko 0:02:03][%emt 0:00:06]}{[%eval -291,11]} 36. dxc5 {[%clk
0:01:58][%clko 0:01:37][%emt 0:00:06]}{[%eval -72,11]} 36... Rxc5 {[%clk
0:01:33][%clko 0:01:58][%emt 0:00:06]}{[%eval -295,11]} 37. Qb6 {[%clk
0:01:50][%clko 0:01:33][%emt 0:00:12]}{[%eval -81,11]} 37... Qc8 {[%clk
0:01:29][%clko 0:01:50][%emt 0:00:06]}{[%eval -317,11]} 38. f4 {[%clk
0:01:24][%clko 0:01:29][%emt 0:00:27]}{[%eval -89,10]} 38... Rxc2 {[%clk
0:01:24][%clko 0:01:24][%emt 0:00:06]}{[%eval -362,10]} 39. Rd3 {[%clk
0:01:16][%clko 0:01:24][%emt 0:00:11]}{[%eval -92,10]} 39... Qf8 {[%clk
0:01:21][%clko 0:01:16][%emt 0:00:06]}{[%eval -377,9]} 40. a6 {[%clk
0:01:14][%clko 0:01:21][%emt 0:00:04]}{[%eval -74,9]} 40... Qc8 {[%clk
0:01:17][%clko 0:01:14][%emt 0:00:06]}{[%eval -302,9]} 41. a7 {[%clk
0:01:13][%clko 0:01:17][%emt 0:00:03]}{[%eval -18,9]} 41... Qa8 {[%clk
0:01:14][%clko 0:01:13][%emt 0:00:05]}{[%eval -291,10]} 42. Rc3 {[%clk
0:01:11][%clko 0:01:14][%emt 0:00:04]}{[%eval 20,9]} 42... Rxc3+ {[%clk
0:01:11][%clko 0:01:11][%emt 0:00:05]}{[%eval -110,11]} 43. Bxc3 {[%clk
0:01:13][%clko 0:01:11]}{[%eval 20,7]} 43... Re7 {[%clk 0:01:08][%clko
0:01:13][%emt 0:00:05]}{[%eval -79,11]} 44. Qd4 {[%clk 0:01:11][%clko
0:01:08][%emt 0:00:03]}{[%eval 22,10]} 44... Bc6 {[%clk 0:01:05][%clko
0:01:11][%emt 0:00:05]}{[%eval -60,11]} 45. Qh8+ {[%clk 0:01:05][%clko
0:01:05][%emt 0:00:08]}{[%eval 8,9]} 45... Kf7 {[%clk 0:01:02][%clko
0:01:05][%emt 0:00:05]}{[%eval -86,11]} 46. Qxh7+ {[%clk 0:00:55][%clko
0:01:02][%emt 0:00:13]}{[%eval 1,11]} 46... Ke8 {[%clk 0:00:59][%clko
0:00:55][%emt 0:00:05]}{[%eval -83,12]} 47. Qg6+ {[%clk 0:00:50][%clko
0:00:59][%emt 0:00:06]}{[%eval 16,11]} 47... Kd7 {[%clk 0:00:57][%clko
0:00:50][%emt 0:00:04]}{[%eval 0,12]} 48. Be5 {[%clk 0:00:49][%clko
0:00:57][%emt 0:00:03]}{[%eval 16,11]} 48... Kc8 {[%clk 0:00:54][%clko
0:00:49][%emt 0:00:04]}{[%eval -79,11]} 49. Qg8+ {[%clk 0:00:37][%clko
0:00:54][%emt 0:00:14]}{[%eval 3,11]} 49... Be8 {[%clk 0:00:52][%clko
0:00:37][%emt 0:00:04]}{[%eval -65,11]} 50. Bd4 {[%clk 0:00:37][%clko
0:00:52][%emt 0:00:02]}{[%eval 3,11]} 50... Qc6 {[%clk 0:00:49][%clko
0:00:37][%emt 0:00:04]}{[%eval 0,11]} 51. Qf8 {[%clk 0:00:37][%clko
0:00:49][%emt 0:00:02]}{[%eval 63,8]} 51... Kd7 {[%clk 0:00:47][%clko
0:00:37][%emt 0:00:04]}{[%eval -20,10]} 52. g5 {[%clk 0:00:38][%clko
0:00:47][%emt 0:00:01]}{[%eval 92,7]} 52... Re4 {[%clk 0:00:45][%clko
0:00:38][%emt 0:00:04]}{[%eval -70,9]} 53. h4 {[%clk 0:00:28][%clko
0:00:45][%emt 0:00:11]}{[%eval 125,8]} 53... Kc7 {[%clk 0:00:43][%clko
0:00:28][%emt 0:00:04]}{[%eval -4,9]} 54. h5 {[%clk 0:00:27][%clko
0:00:43][%emt 0:00:05]}{[%eval 157,7]} 54... Kb7 {[%clk 0:00:41][%clko
0:00:27][%emt 0:00:04]}{[%eval 0,9]} 55. g6 {[%clk 0:00:23][%clko 0:00:41][%emt
0:00:05]}{[%eval 209,8]} 55... Qe6 {[%clk 0:00:40][%clko 0:00:23][%emt
0:00:04]}{[%eval 0,9]} 56. f5 {[%clk 0:00:19][%clko 0:00:40][%emt 0:00:06]}
{[%eval 238,9]} 56... Qc6 {[%clk 0:00:38][%clko 0:00:19][%emt 0:00:04]}{[%eval
-84,8]} 57. Qc5 {[%clk 0:00:18][%clko 0:00:38][%emt 0:00:02]}{[%eval 309,8]}
57... Qxc5 {[%clk 0:00:36][%clko 0:00:18][%emt 0:00:04]}{[%eval -103,10]} 58.
Bxc5 {[%clk 0:00:19][%clko 0:00:36][%emt 0:00:01]}{[%eval 329,10]} 58... Re5
{[%clk 0:00:35][%clko 0:00:19][%emt 0:00:04]}{[%eval 0,10]} 59. Kf4 {[%clk
0:00:19][%clko 0:00:35][%emt 0:00:02]}{[%eval 388,9]} 59... Re4+ {[%clk
0:00:33][%clko 0:00:19][%emt 0:00:03]}{[%eval 0,12]} 60. Kg5 {[%clk
0:00:21][%clko 0:00:33]}{[%eval 388,8]} 60... d4 {[%clk 0:00:32][%clko
0:00:21][%emt 0:00:03]}{[%eval -19,10]} 61. Bxd4 {[%clk 0:00:17][%clko
0:00:32][%emt 0:00:06]}{[%eval 448,10]} 61... Rxd4 {[%clk 0:00:30][%clko
0:00:17][%emt 0:00:03]}{[%eval 0,11]} 62. h6 {[%clk 0:00:14][%clko
0:00:30][%emt 0:00:04]}{[%eval 512,9]} 62... Rd8 {[%clk 0:00:28][%clko
0:00:14][%emt 0:00:03]}{[%eval 418,10]} 63. g7 {[%clk 0:00:14][%clko
0:00:28][%emt 0:00:02]}{[%eval 733,8]} 63... Bf7 {[%clk 0:00:28][%clko
0:00:14][%emt 0:00:03]}{[%eval 423,12]} 64. h7 {[%clk 0:00:12][%clko
0:00:28][%emt 0:00:04]}{[%eval 733,7]} 64... Rg8 {[%clk 0:00:27][%clko
0:00:12][%emt 0:00:03]}{[%eval 423,12]} 65. h8=Q {[%clk 0:00:10][%clko
0:00:27][%emt 0:00:03]}{[%eval 793,7]} 65... Kxa7 {[%clk 0:00:25][%clko
0:00:10][%emt 0:00:03]}{[%eval 537,11]} 66. Kf6 {[%clk 0:00:10][%clko
0:00:25][%emt 0:00:03]}{[%eval 790,8]} 66... Bd5 {[%clk 0:00:24][%clko
0:00:10][%emt 0:00:03]}{[%eval 654,11]} 67. Ke7 {[%clk 0:00:10][%clko
0:00:24][%emt 0:00:02]}{[%eval 850,7]} 67... b4 {[%clk 0:00:23][%clko
0:00:10][%emt 0:00:03]}{[%eval 826,11]} 68. axb4 {[%clk 0:00:10][%clko
0:00:23][%emt 0:00:02]}{[%eval 916,8]} 68... Kb7 {[%clk 0:00:21][%clko
0:00:10][%emt 0:00:03]}{[%eval 1143,11]} 69. f6 {[%clk 0:00:10][%clko
0:00:21][%emt 0:00:02]}{[%eval 1187,6]} 69... Kc6 {[%clk 0:00:21][%clko
0:00:10][%emt 0:00:03]}{[%eval 1950,11]} 70. Qxg8 {[%clk 0:00:10][%clko
0:00:21][%emt 0:00:02]}{[%eval 1981,6]} 70... Bxg8 {[%clk 0:00:21][%clko
0:00:10][%emt 0:00:03]}{[%eval 29965,11]} 71. f7 {[%clk 0:00:09][%clko
0:00:21][%emt 0:00:02]}{[%eval 1983,6]} 71... Bxf7 {[%clk 0:00:20][%clko
0:00:09][%emt 0:00:03]}{[%eval 29986,12]} 72. Kxf7 {[%clk 0:00:09][%clko
0:00:20][%emt 0:00:02]}{[%eval 3267,9]} 1-0
Rybka 3 is quite a bit stronger than the version John Meyer played at knight odds, which had most of the R3 eval but none of the new search. Also, I made a big mistake in not using contempt in that match. I did this because we were getting draw odds, but it was still a big mistake because high contempt would avoid even trades. I think if we tried again with R3 and high contempt (maybe 150 or so), we would not lose every game again, but we would still lose the match 3-1 or 3.5-0.5.
![[au]](/mwf/v12/flags/au.png "Australia")
>odds of the Exchange (Rybka removes White rook from a1, Milov removes Black knight from b8).
I might note (historically) that occasionally (as claimed here) with odds of rook, White also started with the a-pawn at a3 (I guess so that it would be protected). Would this make much difference in the given case? Also, I expect that Rybka can still castle queenside (as per the "standard" of Staunton)?
The placement of the pawn on a3 at rooks odds was only occasionally done, it was never standard (at least in Morphy's time). It would probably decrease the handicap by 25 Elo or so, but we won't ask for this advantage. As for castling with the ghost of the rook on a1, Rybka can't do that anyway of course so we don't even need to talk about this rule; we're certainly not going to create a whole new Rybka version just to accommodate this doubtful rule. And if we did, how would the servers handle this? So forget about this.
![[be]](/mwf/v12/flags/be.png "Belgium")
Why this message don't stay at the top of messages list ?
![[de]](/mwf/v12/flags/de.png "Germany")
Yes, being the next official match and apart from that against a very good opponent, I am really looking forward to this match.
Stick it!
(the last match was already less announced than the matches before I think ... )
Stick it!
(the last match was already less announced than the matches before I think ... )
May be the Rybka team don't figure out anymore this kind of matches are big events. Computer chess fans wait for this for years !
i think the GM should win,,, ill be extremely surprised if rybka can win a strong player like that at odds . of all those odds the exchange in my view will be the hardest for the GM to win because it goes against the patterns of normal chess too much which wont bother rybka at all .
These handicap matches are ludicrous and a waste of Larry's time...time and energy that could be devoted to fixing the bugs in Rybka 3.
I would say that the matches were quite helpful in the development of Rybka 3. Some terms in R3 were implemented as a direct result of the games and/or comments by the GMs about the games. More importantly, the various conditions forced us to do things that proved to be of great value; the new contempt idea was an outgrowth of these matches for example. Also there is considerable interest in them; the last Dzindzi match had over 400 spectators at one point. Finally, I can only fix bugs that are simple numerical errors, like dropping a digit or putting the wrong sign on a value, and although such things have happened they are very rare and have never made it to a release as far as I can remember. Only Vas can fix other bugs.
Now that the match terms are finalized (unless a new sponsor appears), I've added a poll on the result. Thanks to Convekta and to a private sponsor who prefers to remain anonymous for your sponsorship of this match.
![[ar]](/mwf/v12/flags/ar.png "Argentina")
Good, it will be interesting, especially first 2 games!
congrats at getting a 2700 + player , to bad there is handicap but still it will be exciting . if the 2700 player really takes it serious and prepare then my votes for him .unprepared always favors rybka .
Well, I'm pretty sure he is preparing. But probably not by looking at database handicap games; the last ones at Exchange Handicap are 150 years old!
![[it]](/mwf/v12/flags/it.png "Italy")
(Rybka removes White rook from a1, Milov removes Black knight from b8).
But a rook for a knight it is not a great advantage? (too much).
Play a couple of games starting from an opening position of gambit. With strong attack for Milov. Would this be interesting ? Eventually, , an opening position chosen by Milov.
But a rook for a knight it is not a great advantage? (too much).
Play a couple of games starting from an opening position of gambit. With strong attack for Milov. Would this be interesting ? Eventually, , an opening position chosen by Milov.
Although we have little data to go by, I don't think that this handicap is "too much"; I think it's just about right to produce competitive chess that could result in wins for either side. I suppose another way to handicap would be to require that we make a list of unsound gambits or one-sided openings that are evaluated by Rybka 3 at about the same 1.00 score for one side as is the start position of the Exchange Handicap, and either let the human select from them or play each of them once, but to me this is much less "clean" than the Exchange Handicap and it makes the handicap dependent on Rybka's own eval. I have a strong suspicion that this Exchange handicap will produce more interesting and normal chess than the various pawn handicaps have done. It is quite appealing to me, because Rybka at least enjoys a slight initiative as White and due to the extra knight influencing the center to compensate a little for the fact that she is clearly lost in almost any endgame that might arise without a change in the material balance. It really is quite similar to forcing Rybka to play a very unsound gambit. The only problem that I can see with this handicap is that it does not lend itself to modification for players of different levels. If it turns out to be fair for 2700 Milov, then against a 2600 GM we might try Exchange and move, and maybe against lesser players like IM or FM Exchange + "a" pawn (because this pawn's absence is a pure loss without the rook on the file) or Exchange and Move + "a" pawn. But first we have to see what happens with Milov. If we do come out ahead in the Exchange Handicap games with Milov and win the match, I hope we won't still have to hear people claim that Anand or Kramnik or xxx could still compete with Rybka on even terms.
i have a question and try to answer it to your best guess . what would the score be if there was no handicap BUT the GM that will play had rybka 3 and the identical book you will use vs him and 3 weeks to practice .
but knight rook odds is very very interesting ,,, my money is on rybka for that part of it . because i think computers are good if not great in unusual positions ,, ill go on record here to say i cant imagine a human ever competeing with a computer at shuffle chess ,,,, maybe its because im at least 500 to 600 elo weaker at it .
but as far as a human being able to compete vs rybka 3 i cant speculate i did not buy it yet as im way too busy prepare for my tourn . but i think man will always find a way until you have 32 tb then indeed it will take something other than man to compete .
but knight rook odds is very very interesting ,,, my money is on rybka for that part of it . because i think computers are good if not great in unusual positions ,, ill go on record here to say i cant imagine a human ever competeing with a computer at shuffle chess ,,,, maybe its because im at least 500 to 600 elo weaker at it .
but as far as a human being able to compete vs rybka 3 i cant speculate i did not buy it yet as im way too busy prepare for my tourn . but i think man will always find a way until you have 32 tb then indeed it will take something other than man to compete .
Well he does have Rybka 3 and 3 weeks to practice, though we will use a slightly modified/improved R3 partly to avoid having long lines prepared against us. As for identical book, that's a bit undefined. If we use the R3 book (for example) but change priorities, is that still "identical" in your mind? If we can't change anything, he could prepare long lines based on duplicating jor simulating our hardware, and our only defense would be relying on the fact that MP is non-deterministic. I think this would still be enough for us to win such a match, even playing all Black as we are in the non-handicap games, but it would be rather pointless and silly. I mean if the GM starts playing weird moves early on because there is only one book reply, and uses Rybka 3 to extend lines until he finds one that grossly favors him, this is not Rybka vs. GM, this is Rybka vs. Centaur (to a point). As long as you allow us some unpredictibility to avoid exact preparation R3 will win overwhelmingly against any human without a handicap.
I agree with your comments about unusual positions and shuffle chess. But the position with rook on a1 and knight on b8 removed is not nearly as strange as a shuffle chess position; all the pieces that remain on the board are on their normal squares, and castling remains normal (except Rybka can't castle long, making the handicap a bit larger than otherwise). The knight odds match with John Meyer showed that Rybka is rather helpless when totally lost. Being down the Exchange, Rybka will be lost, but not totally lost, so I think that Rybka will be able to demonstrate her strength in this scenario. The result of the handicap games depends very much on Milov. If he plays well enough, he should win the handicap games, as they start with the human in a (probably) winning position. But we don't know just how well he has to play to prove this.
I agree with your comments about unusual positions and shuffle chess. But the position with rook on a1 and knight on b8 removed is not nearly as strange as a shuffle chess position; all the pieces that remain on the board are on their normal squares, and castling remains normal (except Rybka can't castle long, making the handicap a bit larger than otherwise). The knight odds match with John Meyer showed that Rybka is rather helpless when totally lost. Being down the Exchange, Rybka will be lost, but not totally lost, so I think that Rybka will be able to demonstrate her strength in this scenario. The result of the handicap games depends very much on Milov. If he plays well enough, he should win the handicap games, as they start with the human in a (probably) winning position. But we don't know just how well he has to play to prove this.
is there any set date already ??
It is set for the period Sept 14 to Sept 18, only the details of which games on which dates remain to be decided.
A better handicap for Rybka would be for it to play with a 5 or 10 minute time control and be run on a single cpu 32 bit system. The handicap for rybka should be on the time(Rybka would have 5-10 minute per game total time, while the GM would have 360 minutes), ram (64-128mb), cpu speed(1.6GHz-3GHz).
Would the Rybka team consider such conditions to prove computer superiority, Rybka's superiority?
Would the Rybka team consider such conditions to prove computer superiority, Rybka's superiority?
10 min for rybka against 2h/40m for the human ? hell yes humans would be crushed
starting with an equal starting position is too big an advantage for the engine , lol :)
starting with an equal starting position is too big an advantage for the engine , lol :)
This proposal seems good to me... but maybe 6 hours for the GM is not required either, and would probably be too long to be attractive for many kibitzes. But there should be some compensation because the time to think during the opponents time, would be much smaller for the GM, than in normal 40/2h. I would suggest maximal 180+30, and something like 10+5 for Rybka (or 18+3 to make it a 1:10 handicap).
If the Rybka team prefers to run Rybka on very good hardware, and usually with equal time exept once IIRC, I think it is because they want to demonstrate Rybka "at full strength" despite the other conditions. I think that is logical.
Yet, another philosophy was shown when some engines (and Hydra) played against top GMs in Bilbao some years ago, and Fritz was running on an off-the-shelf notebook with a Centrino 1.7 GHz, only. IIRC, Fritz played twice there, in two consecutive years. I would have to look up the details at chessbase.com. But there were no odds whatsoever.
[Event "Man vs Machine 120'/40+60'"]
[Site "Bilbao Esp"]
[Date "2004.10.08"]
[Round "3"]
[White "Fritz"]
[Black "Topalov, Veselin"]
[Result "1-0"]
[ECO "C03"]
[BlackElo "2757"]
[PlyCount "93"]
1. e4 e6 2. d4 d5 3. Nd2 a6 4. Ngf3 Nf6 5. e5 Nfd7 6. Bd3 c5 7. c3 Nc6 8. O-O
g5 9. Bb1 g4 10. Ne1 h5 11. Nb3 a5 12. Nxc5 Nxc5 13. dxc5 Bxc5 14. Nd3 Ba7 15.
Qa4 Bd7 16. Qf4 Bb8 17. Rd1 f5 18. c4 d4 19. Re1 Qe7 20. Bc2 h4 21. Bd2 Bc7 22.
Bd1 Rg8 23. a3 a4 24. f3 gxf3 25. Bxf3 Ba5 26. Bxa5 Rxa5 27. Qd2 Qg5 28. Qf2
Rg7 29. c5 Kf8 30. Rac1 Kg8 31. Kh1 Ra8 32. Rc4 Be8 33. Bd1 Qh6 34. Nf4 Rd8 35.
Rxa4 d3 36. Bb3 Bf7 37. Qe3 Qg5 38. Rd1 Kh7 39. Rd2 Qh6 40. Kg1 Qg5 41. Bc4 Be8
42. Bxe6 Re7 43. Rxd3 Rxd3 44. Qxd3 Nxe5 45. Qxf5+ Qxf5 46. Bxf5+ Kh6 47. Nd5
1-0
If the Rybka team prefers to run Rybka on very good hardware, and usually with equal time exept once IIRC, I think it is because they want to demonstrate Rybka "at full strength" despite the other conditions. I think that is logical.
Yet, another philosophy was shown when some engines (and Hydra) played against top GMs in Bilbao some years ago, and Fritz was running on an off-the-shelf notebook with a Centrino 1.7 GHz, only. IIRC, Fritz played twice there, in two consecutive years. I would have to look up the details at chessbase.com. But there were no odds whatsoever.
[Event "Man vs Machine 120'/40+60'"]
[Site "Bilbao Esp"]
[Date "2004.10.08"]
[Round "3"]
[White "Fritz"]
[Black "Topalov, Veselin"]
[Result "1-0"]
[ECO "C03"]
[BlackElo "2757"]
[PlyCount "93"]
1. e4 e6 2. d4 d5 3. Nd2 a6 4. Ngf3 Nf6 5. e5 Nfd7 6. Bd3 c5 7. c3 Nc6 8. O-O
g5 9. Bb1 g4 10. Ne1 h5 11. Nb3 a5 12. Nxc5 Nxc5 13. dxc5 Bxc5 14. Nd3 Ba7 15.
Qa4 Bd7 16. Qf4 Bb8 17. Rd1 f5 18. c4 d4 19. Re1 Qe7 20. Bc2 h4 21. Bd2 Bc7 22.
Bd1 Rg8 23. a3 a4 24. f3 gxf3 25. Bxf3 Ba5 26. Bxa5 Rxa5 27. Qd2 Qg5 28. Qf2
Rg7 29. c5 Kf8 30. Rac1 Kg8 31. Kh1 Ra8 32. Rc4 Be8 33. Bd1 Qh6 34. Nf4 Rd8 35.
Rxa4 d3 36. Bb3 Bf7 37. Qe3 Qg5 38. Rd1 Kh7 39. Rd2 Qh6 40. Kg1 Qg5 41. Bc4 Be8
42. Bxe6 Re7 43. Rxd3 Rxd3 44. Qxd3 Nxe5 45. Qxf5+ Qxf5 46. Bxf5+ Kh6 47. Nd5
1-0
It should be regulated, fully observed by FIDE arbiters on each side/each playing area, so they really play by the rules.
Fritz back then still played on a powerful hardware, considering that centrino on laptop cannot be compared with 3GHz pentium 4(P4 is slower), the speed count was misleading. Plus the RAM was quite big.
Handicap should be on the use of old machines, say an 800MHz 32 bit single cpu with physical RAM of 128-256MB, ones that you usually see being used as home PCs.
If Rybka can do well on cheap hardware and remain unbeatable, then you can really say that it did not depend to a large extent on its hardware.
Fritz back then still played on a powerful hardware, considering that centrino on laptop cannot be compared with 3GHz pentium 4(P4 is slower), the speed count was misleading. Plus the RAM was quite big.
Handicap should be on the use of old machines, say an 800MHz 32 bit single cpu with physical RAM of 128-256MB, ones that you usually see being used as home PCs.
If Rybka can do well on cheap hardware and remain unbeatable, then you can really say that it did not depend to a large extent on its hardware.
> Handicap should be on the use of old machines, say an 800MHz 32 bit single cpu with physical RAM of 128-256MB, ones that you usually see being used as home PCs.
Yes, it is interesting to see R3 working on such an old hardware.
But:
Do you really think, that today a 800MHz-CPU can be usually seen on home PCs?
I can't imagine that in Europe an northern America.
P4 with 2.6 GHz is the weakest hardware I can detect all around me.
I think: 800 MHz you can only find in EDV museums and the rooms of nostalgic EDV-gatherers, 2nd (no, not really) and 3rd and 4th PCs, today.
Quap
I am still running an AthlonXP 1900, unfortunately I need to underclock it to ~1200...1400 MHz although I have giant fans built in on that comp. Usually, all engines run fine with the exception that under Windows 98SE which I use at it, some engines cannot use more than 255 MB and if they use 2^x values only, max. 128 MB. That is due to a programming flaw which ignores Win98 requirements. There are other engines which can use 256 MB and more without problem. Since my tests usally are at short to medium time controls, that is not a problem at all (Win98 is the bigger problem).
One reason that I still use this old computer for testing is, that my newer computer is a D945 which I use for the internet access most of the time, and also that one gets very loud due to the cooling fans, after longer full cpu load.
With unlimited money, I'd have a computer's hall with a couple of 16-core comps, air condition and dust-free conditions, etc.etc. :-D but in practice of life, money is limited and there are several priorities.
One reason that I still use this old computer for testing is, that my newer computer is a D945 which I use for the internet access most of the time, and also that one gets very loud due to the cooling fans, after longer full cpu load.
With unlimited money, I'd have a computer's hall with a couple of 16-core comps, air condition and dust-free conditions, etc.etc. :-D but in practice of life, money is limited and there are several priorities.
![[fr]](/mwf/v12/flags/fr.png "France")
> One reason that I still use this old computer for testing is,
... that I am cheapo... a miser... somebody who has such a low opinion of myself that I value the time I spend on chess as next to zero...
Come on, be a Mensch, spend 500 euros on your best hobby! :-p
800MHz on a Pentium you have on Pentium III, started to be built in 1999
your 'old computer' AthlonXP 1900 came twio years later.
I assume, that today 800MHz-PCs are very, very rare.
Germany, France, USA...
But how is the situation in Romania, in Brasilia or Namibia.
OK, maybe, there even 800MHz PCs are running today, carrying W98, Linux or maybe even XP.
Quap
your 'old computer' AthlonXP 1900 came twio years later.
I assume, that today 800MHz-PCs are very, very rare.
Germany, France, USA...
But how is the situation in Romania, in Brasilia or Namibia.
OK, maybe, there even 800MHz PCs are running today, carrying W98, Linux or maybe even XP.
Quap
"With unlimited money, I'd have a computer's hall with a couple of 16-core comps, air condition and dust-free conditions, etc.etc. :-D but in practice of life, money is limited and there are several priorities."
I agree with you 100% and I share the same sentiment.
I agree with you 100% and I share the same sentiment.
286-386MHz PCs are present in third world/developing countries like the Philippines.
That is what I call true handicap.
That is what I call true handicap.
> 286-386MHz PCs are present in third world/developing countries like the Philippines.
> That is what I call true handicap.
This MHz-values are in the range of a Pentium II. (1997/1998)
I think, it would be very interesting to see, what Rybka can reach on such hardware playing against her actual follower playing on modern Quads.
But maybe you meant not MHz-values but PCs with Intel 286 (from 1982) and Intel 386 (from 1985).
Are also those machines used in counties like the Philippines today?
(These special Numbers you told of are catching my eyes)
OK, that Intel-286 can't host the modern Software like Rybka.
Which ist the strongest chess-software runnable on a such a Intel 286
But maybe the Intel-386 (or even the 386SX) can do it.
Has anybody tried to run R3 or R2 on such a machine, maybe under W98?
I think, this OS can run on a Intel 386, right? Is R3 or R2 content with W98?
Quap
![[fi]](/mwf/v12/flags/fi.png "Finland")
Intel 386 + Windows 98 should be sufficient, provided you have enough RAM.
Well, we considered both the idea of running on obsolete cheap hardware and the idea of giving huge time handicaps (with ponder off as a further handicap), but not the idea of doing both. It doesn't make sense to give huge time handicaps with ponder on, because then the player might be tempted to play second-best moves to avoid the predicted move, and because the size of the handicap depends on the frequency of ponder hits, which in turn depends on the player's style. With ponder off, I think we could give sixty to one time odds (like five hours to five minutes) on my octal computer, but certainly not on a 32 bit single cpu. In general we are more interested in showing how well Rybka can play with equal time and on good but affordable hardware, so we prefer on-the-board handicaps like material, but if a sponsor requests time handicap or obsolete hardware we would oblige.
![[il]](/mwf/v12/flags/il.png "Israel")
"With ponder off, I think we could give sixty to one time odds (like five hours to five minutes) on my octal computer, but certainly not on a 32 bit single cpu."
I see no reason to be certain about it considering the fact that you do not play against a top GM.
I think that rybka is favourite even with 60:1 time handicap on a 32 bit single cpu.
Rybka3 32 bit may have more than 3100 CCRL rating
see http://computerchess.org.uk/ccrl/404.live/rating_list_all.html
If you reduce 400-500 elo for the time handicap then you get 2600-2700 CCRL rating for rybka that is probably better than 2550 fide rating
Edit:I thought about the match against another GM and now I see that this subject is about the match against Vadim milov.
Uri
I see no reason to be certain about it considering the fact that you do not play against a top GM.
I think that rybka is favourite even with 60:1 time handicap on a 32 bit single cpu.
Rybka3 32 bit may have more than 3100 CCRL rating
see http://computerchess.org.uk/ccrl/404.live/rating_list_all.html
If you reduce 400-500 elo for the time handicap then you get 2600-2700 CCRL rating for rybka that is probably better than 2550 fide rating
Edit:I thought about the match against another GM and now I see that this subject is about the match against Vadim milov.
Uri
One other point: I still believe that the CCRL and to a lesser degree CEGT ratings somewhat overrate today's engines in human terms, not because they do anything seriously wrong, but in part because humans do not play from short opening books with randomized openings but often prepare lines all the way to the endgame. I believe that if Rybka is rated 3200 on some hardware by CEGT (I'll use the more conservative rating list here), she would probably be able to justify this rating (at least approximately) in a tournament with the top six human players if they used CEGT opening books and reversed colors, but playing with no restrictions I think the 3200 rating would drop at least to 3100. The opening is the key here. If we can avoid deep opening preparation when we have Black without landing in a nearly lost position, we can score very well against top humans, but that isn't so easy to do, so top humans always have some hope for a draw when they have White.
> If we can avoid deep opening preparation when we have Black without landing in a nearly lost position, we can score very well against top humans, but that isn't so easy to do
Why is it not easy? If a wide and flat opening book is used I can not imagine that any human has all lines in his head, or does he? I'm only a weak player.
> I think the 3200 rating would drop at least to 3100. The opening is the key here.
I remember you sad that rybka performs better against related opponents. Contrary I would conclude that the rybka improvement is less valuable against unrelated and very dissimilar opponents like humans. I belief this is another reason why the rating lists could be inflated in terms of human opponents.
Let's assume two entities A and B figthing against each other in any game not necessarily chess. Both have completely different strengths and weaknesses. A is very good in situations of type X but very bad in ones of type Y. B vice versa. X and Y have the same relevance and occurence. The performance of A against B is 50%. A improves both abilities X and Y each with 100 ELO. That means A would perform 100 ELO better against itself, but curiosly nearly 0 ELO better (50%) against B. Despite of the improvements the result depends mainly on the occurence of X and Y that is 50%.
That means an overall improvement is less valuable against an opponent with completely different strengths and weaknesses. Furter, a partial improvement of an element where the opponent is much stronger or much weaker is less valuable than an improvement where the opponent has the same strength.
Against a top-five human player it is very easy to get a nearly lost position as Black even playing standard openings. They know their theory very well, and only a small percentage of plausible openings are really fully playable for Black. Rybka could probably draw most of those bad positions, but winning them might be nearly impossible. Your argument about strengths and weaknesses is a very good explanation of my long-time contention that the more dissimilar two entities are, the less decisive a given improvement in strength will be. As you say, humans and engines are especially dissimilar and so this is another reason (besides the openings) why the 3200+ ratings for Rybka are not likely to be seen against human opposition. Maybe loping about 1/3 of the excess over the reference engine would give a good estimate of human ratings. So if Rybka 3 on good hardware is 3200 CEGT while the reference engine is 2750, then 3050 might be a realistic figure against the top human players.
Another area being considered is this pondering/permanent brain...Does it really work?
Ponder on does not really matter in my opinion, Rybka or another engine for that matter doesn't fully utilize pondering as if its own time. Gains are minor, too minute to be appreciated. Humans can still play good moves, here are examples on other softwares:
http://chessexchange.com/forum/viewtopic.php?t=1495&highlight=defeat
http://chessexchange.com/forum/viewtopic.php?t=1516&highlight=defeat
http://chessexchange.com/forum/viewtopic.php?t=2427&highlight=defeat
Ponder on does not really matter in my opinion, Rybka or another engine for that matter doesn't fully utilize pondering as if its own time. Gains are minor, too minute to be appreciated. Humans can still play good moves, here are examples on other softwares:
http://chessexchange.com/forum/viewtopic.php?t=1495&highlight=defeat
http://chessexchange.com/forum/viewtopic.php?t=1516&highlight=defeat
http://chessexchange.com/forum/viewtopic.php?t=2427&highlight=defeat
In normal play (with equal time), ponder on is not a big deal; I would guess it is worth about 15 Elo points. Surely this could be improved, but probably not by very much, maybe five to ten Elo points. So it's not a high priority. Ponder on does make a big difference in games with a huge time handicap like in our latest match with GM Perelshteyn, that's why we turned it off.
Using time for waiting for the next move only by "ponder on" seems to unnecessary waste time and resources.
Wouldn’t it be helpful to invent some other procedures to use hard- and software-power in a more productive way?
But which procedures?
Wouldn’t it be helpful to invent some other procedures to use hard- and software-power in a more productive way?
But which procedures?
Yes, surely better use could be made of ponder time. But the likely gains are too small to justify working on this now, when there are still much larger improvements to be made.
> But the likely gains are too small to justify working on this now, when there are still much larger improvements to be made.
That's surely true and you and Vas proofed, that you know well how to prioritize.
(Nevertheless I am wondering if it will help somehow to let Rybka make some Monte Carlo researches from the perspecitve of the opponents side while waiting, taking the result somehow in account. But that's only a vague idea, too vague to keep track on, at least in short time controls.)
Yeah, like it might not need ponder time if the quality of the moves even at bullet or blitz time controls is astonishing and can match the moves made at long time controls.
The software that can deliver results in the least amount of time and effort wins.
The software that can deliver results in the least amount of time and effort wins.
Yes, ponder on was not able to save the native chessbase engines from being beaten in the links given above.
If I may suggest, ponder on should be used by Rybka in a different way, something like she'd use it as her own normal time and not just predict moves with so little gain.
If I may suggest, ponder on should be used by Rybka in a different way, something like she'd use it as her own normal time and not just predict moves with so little gain.
It would be possible to have Rybka consider all legal replies and not just the "best" one, but I doubt that this would actually be an improvement. What might be an improvement is to do a short search to select all moves within some narrow window of the best one, and ponder on all of them.
![[pl]](/mwf/v12/flags/pl.png "Poland")
One other issue is that the GUI expects analysis of the move which the engine self-reports as the ponder move.
Vas
Vas
it doesn't matter as long as rybka plays the best moves all the time.
Yeah, I mean the best possible one, within the time it's pondering.
The best way to use ponder is for rybka to see it as its own time and perform search as if its having its own time.
The best way to use ponder is for rybka to see it as its own time and perform search as if its having its own time.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill